Most candidates who "build a driver tree" are drawing an issue tree with nicer labels. A real driver tree is an arithmetic identity: every child multiplies or adds into its parent, so a change in any leaf reconstructs the headline number. After this piece you will know when that discipline beats an issue tree or a decision tree, how to spot the classic leaks (non-MECE branches, stocks mixed with flows), and how to run the math on a live prompt.
The test the interviewer is actually running
McKinsey's interviewing page says the problem-solving interview checks how you structure ambiguous challenges, work facts and data, and form a recommendation. In a profitability or KPI case, the structure they reward is not a list of themes. It is a decomposition you can calculate.
I stop listening when a candidate writes "Revenue | Costs | Customers | Competition" under profit. Customers already sit inside revenue. Competition is a cause, not a math input. The sheet is listening for a tree that rebuilds: Profit = Revenue − Cost; Revenue = Price × Volume; and so on. If the branches do not add or multiply back to the parent, you have not built a driver tree.
Arithmetic first: what a driver tree is
Put the metric the client cares about at the top. Under it, place only the inputs that combine by a defined operation to produce that metric. Under each input, repeat until the leaves are measurable.
For a simple P&L:
- Profit = Revenue − Total cost
- Revenue = Price × Volume
- Volume = Customers × Purchases per customer (or units sold, or load-miles — pick one definition and stick to it)
- Total cost = Fixed cost + Variable cost
- Variable cost = Unit variable cost × Volume
The distinctive property is reconstructability. If you change volume by 10% and hold price and unit cost, you can state the revenue and variable-cost moves before any exhibit arrives. That is why the same logic shows up in real engagements: it turns diagnosis into arithmetic instead of storytelling.
MECE still matters, but for a driver tree the MECE test is stronger than "no overlap of topics." Branches must be mutually exclusive as math inputs, and collectively exhaustive in the sense that they rebuild the parent exactly. The MECE principle is the checklist; the equation is the proof.
Where the idea came from: DuPont's identity
The driver tree is older than case interviews. In 1914, F. Donaldson Brown at DuPont built a return measure that combined earnings, working capital, and plant investment into one rate — what later became the DuPont method for return on investment. The Hagley Museum's account of Brown's work places that formulation in the Treasurer's Office as DuPont diversified and needed a single performance gauge across departments.
The modern three-factor form taught in finance — often called the DuPont identity — writes return on equity as a product:
ROE = (Net income / Sales) × (Sales / Assets) × (Assets / Equity)
Margin × turnover × leverage. Multiply the three and the intermediate terms cancel; you get net income over equity again. OpenStax's Principles of Finance states the same DuPont method algebra, and a CFA Institute Inside Investing note walks the identity on real filings. Change one factor and you can see which lever moved ROE. That is the entire point of a driver tree: inputs that, with the right operation, rebuild the number you started with.
If your branches do not multiply or add back to the parent, you have an issue tree wearing a driver-tree costume.
Three trees, three jobs
Candidates blur three tools that look similar on a whiteboard and do different work.
| Tool | What the branches are | Operation | Use when |
|---|---|---|---|
| Driver tree | Inputs to a metric | Multiply or add (identity) | Why a number moved, or how to move it |
| Issue tree | Questions or hypotheses | Categorical, MECE by judgment | Frame an ambiguous problem before you have data |
| Decision tree | Choices and uncertain outcomes | Expected value across chance nodes | Choose among options under uncertainty |
An issue tree asks open questions: "Is the profit drop revenue-led or cost-led?" "Is volume the problem or price?" It does not claim that answering those questions reconstitutes the dollar figure. A driver tree does.
A decision tree is a different animal again. In decision analysis, square nodes are choices you control; circular nodes are chance outcomes with probabilities; you fold the tree back by expected value. Brunel University's operations-research note on decision trees is a durable primer on that notation. Use a decision tree when the case is "build the plant / wait / exit," not when the case is "explain the $1.8M miss." For the choice-under-uncertainty version in interview language, see the sibling guide on the decision tree framework.
In a typical first round, the sequence is issue tree then driver tree: frame the question, then quantify the lever. Skip the framing and you dig into the wrong identity. Skip the arithmetic and you never prove impact.
Worked example: Meridian Fasteners' $1.8M miss
Invented company, round numbers you can redo on paper.
Meridian Fasteners distributes industrial bolts to OEMs. Last year profit was $6.0M. This year it is $4.2M. The CEO wants the lever, not a strategy essay.
Base year (ties out):
| Line | Value |
|---|---|
| Units sold | 2,400,000 |
| Average selling price | $12.50 |
| Revenue | 2,400,000 × $12.50 = $30,000,000 |
| Unit COGS | $7.50 |
| COGS | 2,400,000 × $7.50 = $18,000,000 |
| Operating expense (fixed for this cut) | $6,000,000 |
| Profit | $30M − $18M − $6M = $6,000,000 |
Driver tree:
Profit = Revenue − COGS − OpEx
Revenue = Price × Units
COGS = Unit COGS × Units
This year (one change isolated at a time):
- Units fell 10% to 2,160,000; price held at $12.50; unit COGS held at $7.50; OpEx held at $6M.
- Revenue = 2,160,000 × $12.50 = $27,000,000 (−$3.0M)
- COGS = 2,160,000 × $7.50 = $16,200,000 (−$1.8M)
- Profit = $27.0M − $16.2M − $6.0M = $4,800,000 (−$1.2M from volume alone)
Volume explains $1.2M of the $1.8M miss. You still need $0.6M.
- Hold the new volume, raise unit COGS to $7.80 (a 4% cost creep), keep price and OpEx:
- COGS = 2,160,000 × $7.80 = $16,848,000
- Profit = $27.0M − $16.848M − $6.0M = $4,152,000
Close enough to $4.2M for interview rounding. Synthesis in one line: about two-thirds of the miss is volume; the rest is unit cost; price and OpEx did not move. You did not guess "maybe competition." You proved it on the identity.
That is what the interviewer is scoring: not the diagram, the trail from leaf to parent.
Ned's rule. If you cannot write the operator between a node and its parent (+ or ×), erase the node. Causes and themes belong on an issue tree; only math inputs belong on a driver tree.
Classic errors that cost the case
Non-MECE branches. "Revenue, cost, and customers" double-counts. "Price, volume, mix, and discounts" can also double-count if discounts already sit inside net price. Pick one price definition (list, net, or realized) and one volume definition, then stick to them.
Stocks mixed with flows. Balance-sheet stocks (inventory, receivables, headcount) do not sit next to period flows (revenue this year, COGS this year) without a bridging equation. Inventory affects COGS through units sold and unit cost, or through working-capital turns in a DuPont-style ROE tree — not as a sibling of "Revenue" under Profit. Mixing a stock with a flow is how candidates invent fake levers that never reconcile.
Causes dressed as drivers. "Competition," "macro," and "brand" can explain why a driver moved. They are not themselves drivers unless you define a measurable link (share × market size, for example). Put them on the issue tree; put share and market size on the driver tree.
Wrong tree for the prompt. Market entry and "should we acquire" cases are usually issue trees first, sometimes decision trees if options and uncertainty dominate. Forcing Profit = Revenue − Cost onto every prompt is a tell that you memorized a template.
For profitability framing more broadly, see profitability cases. For the wider structure toolkit, see how to pick a case structure.
Practice this today
Take any KPI — profit, contribution margin, revenue per store, ROE — and write only equations until the leaves are observable numbers. Then change one leaf by 10% and recompute the parent. If you cannot finish in five minutes, the tree is still categorical.
Drill the same habit under time: CoachNed's structure drills score whether your branches stay MECE and useful. When you want to assemble a full case structure on a blank page, use the case-structure builder. Everything on CoachNed is open for seven days with no card; then $120 for a recruiting season or $49 a month.
Frequently asked questions
What is a driver tree in a case interview?
A driver tree decomposes one measurable metric into inputs connected by addition or multiplication, so every branch rebuilds its parent. It is the quantitative cousin of an issue tree: same visual shape, different rule.
How is a driver tree different from an issue tree?
An issue tree organizes questions or hypotheses into MECE categories. A driver tree organizes math inputs into an identity. Use the issue tree to decide what to investigate; use the driver tree to prove which lever moved the number.
How is a driver tree different from a decision tree?
A decision tree maps choices and chance outcomes to pick an option under uncertainty, usually with probabilities and expected values. A driver tree explains or forecasts a single metric. Do not put probability nodes on a profitability identity.
Where did driver trees come from?
The practical ancestor is DuPont's early-20th-century ROI decomposition, associated with Donaldson Brown around 1914, later generalized in finance as the DuPont identity for ROE. Consulting interviews borrow the same reconstruct-the-metric discipline.
What is the most common driver tree mistake?
Drawing overlapping buckets (revenue, cost, and customers) or mixing balance-sheet stocks with P&L flows so the tree never ties out. If the math does not close, the interviewer will notice before you finish the recommendation.
Sources
- Interviewing at McKinsey — firm description of the problem-solving interview and what it evaluates. Checked 2026-09-24.
- The Father of ROI: Donaldson Brown | Hagley — primary institutional history of Brown's 1914 ROI formulation at DuPont. Checked 2026-09-24.
- Profitability Ratios and the DuPont Method — OpenStax — textbook statement of ROE = margin × turnover × equity multiplier. Checked 2026-09-24.
- How Much Does Apple Make? A DuPont Analysis — CFA Institute — worked DuPont identity on public filings. Checked 2026-09-24.
- Decision Trees — Brunel University OR notes — decision nodes, chance nodes, and expected-value foldback. Checked 2026-09-24.
CoachNed is not affiliated with McKinsey, BCG, Bain, Deloitte, DuPont, or any firm named above.